5
$\begingroup$

Suppose $(f_n)$ is a sequence of probability density functions on $\mathbb{R}$ such that \begin{align*} f_n &\leq M \text{ for all }n \\ f_n(x) &= 0 \text{ for all } |x| > 1 \end{align*} Suppose also that there is a Borel probability measure $\mu$ on $\mathbb{R}$ such that $$ \lim_{n \rightarrow \infty} \int gf_n dx = \int g d\mu \quad \forall g \in C_b(\mathbb{R}), $$ where $C_b(\mathbb{R})$ is the set of all bounded, continuous, complex-valued functions on $\mathbb{R}$. (In other words, suppose the measures $\mu_n$ defined by $d\mu_n = f_n dx$ converge to $\mu$ weakly.)

Question: Is $\mu$ absolutely continuous?

$\endgroup$
4
$\begingroup$

Yes, it is.

Step 1 We have $\mu(A)=0$ for any $A \in \mathcal{B}(\mathbb{R})$ such that $$d(A,[-1,1]) := \inf_{\substack{x \in A \\ y \in [-1,1]}} |x-y|>0.$$ Indeed: Pick a cut-off function $\chi_k \in C_b$ satisfying $\chi(x)=0$ for $|x| \leq 1$, $\chi(x)=1$ for $|x| > 1+1/k$. Since $f_n(x)=0$ for $x \in \mathbb{R} \backslash [-1,1]$ the weak convergence implies

$$\int \chi_k(x) \, d\mu(x) = \lim_{n \to \infty} \int \chi_k(x) \cdot f_n(x) \, dx =0$$

Hence,

$$\mu(A) = \int 1_A \, d\mu \leq \int \chi_k(x) \, d\mu(x) =0$$

for any $K$ such that $d(A,[-1,1]) > 1/k$.

Step 2 By the Radon-Nikodym Theorem, $\mu$ is absolutely continous with respect to the Lebesgue measure $\lambda$, if and only if, $\lambda(A)=0$ implies $\mu(A)=0$ for each $A \in \mathcal{B}(\mathbb{R})$. Since the first part of the proof implies in particular $\mu(\mathbb{R} \backslash [-2,2])=0$, it suffices to show $\lambda(A)=0$ implies $\mu(A)=0$ for $A \subseteq [-2,2]$.

Let $A \subseteq [-2,2]$ be open. We define

$$A_k := \{x \in \mathbb{R}; |x| \leq 2; d(x,A^c) \geq 1/k\}, \qquad k \in \mathbb{N}$$

Obviously, $A_k$ is compact and $A_k \uparrow A$. It is not difficult to show that there exists a sequence $(g_k)_{k \in \mathbb{N}} \subseteq C_C$ such that $1_{A_k} \leq g_k \leq 1_A$ and $g_k \uparrow 1_A$. Therefore, we find by the monotone convergence theorem and the weak convergence

$$\mu(A) = \sup_{k \in \mathbb{N}} \int g_k \, d\mu = \sup_{k \in \mathbb{N}} \limsup_{n \to \infty} \int g_k(x) \cdot f_n(x) \, dx$$

Since $0 \leq f_n \leq M$ and $g_k \uparrow 1_A$, we find

$$\mu(A) \leq \limsup_{n \to \infty} \int 1_A(x) \cdot f_n(x) \, dx \leq M \cdot \lambda(A) \tag{1}$$

Recall that a finite measure on $(\mathbb{R},\mathcal{B}(\mathbb{R}))$ is outer regular, i.e.

$$\lambda(B) = \inf\{\lambda(A); A \supseteq B, A \, \text{open}\} \tag{2}$$

By $(1)$ we find that for any $B \in \mathcal{B}(\mathbb{R})$ and $A \supseteq B$ open,

$$\mu(B) \leq \mu(A) \leq M \cdot \lambda(A)$$

Consequently, $(2)$ entails $\mu(B)=0$ for any set $B$ satisfying $\lambda(B)=0$.

Remark Note that the assumption on the uniform boundedness is crucial, see for instance this question.

$\endgroup$
  • $\begingroup$ Hello ! Nice answer. Would you have a reference for this proof, please (I mean a textbook or a monograph where such a proof and similar results are shown)? $\endgroup$ – MoebiusCorzer May 14 '17 at 11:58
  • 1
    $\begingroup$ @MoebiusCorzer Sorry, I don't have a reference for the result or the proof. Billingsley ("Convergence of probability measures") discusses weak convergence quite thoroughly, but I'm not sure whether this result is in there. $\endgroup$ – saz May 14 '17 at 12:55
  • $\begingroup$ Thanks for your prompt answer. I have checked Billingsley's book but haven't found a proof. It is not a problem, I will adapt it to my problem. $\endgroup$ – MoebiusCorzer May 14 '17 at 12:56
2
$\begingroup$

One may also prove this via functional analysis. Indeed, for $g \in C^{\infty}([-1,1])$, let us define $$Q(g):= \int g d\mu.$$ By the given condition and Cauchy-Schwarz, we have that $$|Q(g)| = \lim_{n \to \infty} \big| \langle f_n,g \rangle_{L^2} \big|\leq \limsup_{n \to \infty} \|f_n\|_{L^2} \|g\|_{L^2} \leq 2^{1/2}M \|g\|_{L^2},$$ so that $Q$ can be extended to a bounded linear operator from $L^2[-1,1] \to \Bbb R$. By the Riesz representation theorem, we know any bounded linear functional on $L^2$ is given by $\langle f, \cdot \rangle$ for some $f$. Thus there exists $f \in L^2[-1,1]$ such that for all $g \in L^2[-1,1]$ we have $$Q(g) = \int fg$$which then implies that $\mu$ is absolutely continuous with density $f$.

$\endgroup$
  • $\begingroup$ I really liked your approach. Could you please help me understand the last sentence though? Why $\mu$ is absolutely continuous with density $f$? $\endgroup$ – kaithkolesidou Feb 20 at 11:23
  • $\begingroup$ @kaithkolesidou the point is, if $\int gf \; dx = \int g \;d\mu $ for all continuous $g$, then one necessarily has that the measure $f\;dx$ must agree with $\mu$, since measures are determined by their integrals against continuous test functions. $\endgroup$ – Shalop Feb 20 at 15:16
  • $\begingroup$ yes I can see that. But this is equivalent to absolutely continuity because there exists a measurable $f$ such that $\mu=fdx$? I thought only the opposite direction is true $\endgroup$ – kaithkolesidou Feb 20 at 15:22
  • $\begingroup$ @kaithkolesidou I think your understanding of absolute continuity is backwards. $\endgroup$ – Shalop Feb 20 at 21:16
  • $\begingroup$ As I see it, $\mu$ is $\sigma$-finite on $[-1,1]$ and thus by $d\mu=fdx$ we can deduce by Rdon-Nikodym theorem the absolutely continuity of $\mu$. Please correct me if I 'm wrong because this is what I'm missing... However thanks a lot already for your time! $\endgroup$ – kaithkolesidou Feb 21 at 10:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.